TLFIG-Transformer: Trend-Oriented Linear Fuzzy Information Granulation-Based Transformer Model for Long-Term Time-Series Forecasting

Long-term time series forecasting holds significant practical value in various fields. However, time series data exhibit complex dynamic characteristics, including trends, seasonality, and noise. These features pose substantial challenges for long-term forecasting. This paper proposes a trend-oriented fuzzy information granulation method and combines it with Transformers to construct a long-term time series forecasting model. By performing decomposition on the time series, the model can extract core trend information and apply fuzzy granulation to it, reducing the influence of other interfering factors on the forecasting process. To better capture long-term dependencies in time series data, this paper designs multiple Transformer modules. Some of these modules are designed to process trend information at the granular level, while others model additional features at the numerical level. This architecture adopts a direct multi-step forecasting strategy, avoiding the error accumulation issue commonly found in traditional autoregressive models. Experimental results demonstrate that the proposed model offers significant advantages in improving forecasting accuracy, handling complex dynamic changes, and enhancing computational efficiency. By integrating trend information granulation with deep learning models, this paper presents an innovative and efficient solution for long-term time series forecasting tasks.